Anthropic Opens Its Doors to Independent AI Evaluators

On October 15, 2026, Anthropic shared how it is working with independent, third party evaluators to review its AI models and safety practices.

The biggest step: six people from METR, a nonprofit that evaluates frontier AI, now have access similar to Anthropic's own risk teams, including internal tools, systems and office space. METR will publish its results, and Anthropic cannot veto what it publishes. Anthropic can only ask to redact a few predefined areas, such as customer data and model architecture details.

SecureBio is reviewing the chemical and biological risk chapter of Anthropic's latest Risk Report, and Anthropic is also working with Accenture and Faculty, with early talks with Hugging Face and others.

Anthropic compares this to bank supervision, where examiners work inside the largest banks. Good evaluators need expertise, independence, real access and the freedom to publish. Anthropic says every frontier AI company should do the same.

Why it matters: trust in AI is built with transparency and accountability, not only with better models.